ICR-DETR:面向工业领域的无损检测模型OA

Accurate and efficient detection of steel surface defects is very important for industrial quality control.Although the RT-DETR model provides a good balance between speed and accuracy,its fixed feature extraction method and detection accuracy are still limited when dealing with complex and fine-grained defects.To this end,this paper proposes an improved detection algorithm based on RT-DETR-ICR-DETR.Firstly,UniRepLKNet is used as the backbone network to significantly improve the model's perception of metal surface features.Secondly,the LWN-CS module is introduced into the neck network,which combines the learnable wavelet transform and channel shuffle mechanism to effectively enhance the high-frequency detail modeling and feature fusion capabilities.Finally,the Shape-WIoU loss function is designed,and the non-monotonic sample weighting mechanism of WIoU is fused with Shape-IoU to comprehensively model the shape feature and scale correlation of the bounding box,so as to improve the positioning accuracy and robustness.The experiment is carried out on the open steel surface defect dataset NEU-DET.The results show that ICR-DETR achieves 77.92%precision,75.71%recall and 78.42%mAP,which is superior to the existing mainstream detection algorithms.In addition,in order to verify the generalization ability of the model,we conduct tests on the self-built sliding contact line shedding dataset ICRFD and heavy truck crack dataset FDMPI to verify its practicability and robustness in complex industrial environments.The experimental results show that the proposed ICR-DETR can effectively support high-precision defect detection tasks in complex industrial scenarios.